On estimating the Carbon Footprint of Computational Fluid Dynamics

Since their invention, the capability of computational machines to simulate detailed physical phenomenon has grown by many orders of magnitude. These innovations have meant that one generation's main frame computer simulations are next generation's workstation simulations and are next generation's smartphone simulations. Historically unachievable simulations, flow at high Reynolds number has started to become feasible. Yet, there is a cost to the growing reliance on large-scale simulations, not just in time and money, but carbon equivalent produced from the electricity consumption. This cost is connected to the number of core-hours of each simulation and is especially onerous for direct numerical simulation. In this talk, we develop estimates of the carbon footprint associated with this paradigm by examining simulations of canonical turbulent flows (isotropic, channel, boundary layer, etc.). We propose that the developed scalings should be used in conjunction with other factors used in tackling a new problem. While high resolution simulations can be carbon heavy, they offer the promise of accurate reduced order model development which are themselves lower sources of carbon. Finally, the adopted scalings may serve as motivation to invest in funding for renewable energy sources which reduce the impact of studying physics.